TY - JOUR
T1 - BRB Prediction with Customized Attributes Weights and Tradeoff Analysis for Concurrent Fault Diagnosis
AU - Chang, Leilei
AU - Xu, Xiaojian
AU - Liu, Zhun Ga
AU - Qian, Bin
AU - Xu, Xiaobin
AU - Chen, Yu Wang
N1 - Publisher Copyright:
© 2007-2012 IEEE.
PY - 2021/3
Y1 - 2021/3
N2 - Comprised of multiple belief rules that are close to human knowledge representation, the belief rule base (BRB) has the ability to make full use of both data obtained from historic records and experts' knowledge under uncertainty. When applied in fault diagnosis, the final diagnostic results are inferred by first activating the matching degrees and activation weights, and then integrating the activated rules with the evidential reasoning algorithm. In this study, a novel BRB prediction approach is proposed for concurrent fault diagnosis. In the proposed approach, the indicative factors are modeled as attributes with customized attribute weights to represent their relevance to a known system fault and vice versa. Second, a tradeoff analysis is conducted by comparing the beliefs for different faults. The diagnosed faults with bigger beliefs within a predetermined valve would be considered as concurrent faults. To further improve the prediction accuracy, an optimization approach with the differential evolutionary algorithm as the optimization engine is designed with model constraints reflecting the complex correlations between the indicative factors and concurrent faults. A marine diesel engine fault diagnosis case is studied for validation. Case study results show that: first, both single and concurrent faults can be diagnosed with high accuracy by the proposed BRB prediction approach; second, the BRBs with optimized attributes weights has shown superior performance than with equal and fixed attribute weights; and third, the new BRB prediction approach outperforms the previous approach with multiple sub-BRBs as well as other approaches to achieve a higher modeling accuracy.
AB - Comprised of multiple belief rules that are close to human knowledge representation, the belief rule base (BRB) has the ability to make full use of both data obtained from historic records and experts' knowledge under uncertainty. When applied in fault diagnosis, the final diagnostic results are inferred by first activating the matching degrees and activation weights, and then integrating the activated rules with the evidential reasoning algorithm. In this study, a novel BRB prediction approach is proposed for concurrent fault diagnosis. In the proposed approach, the indicative factors are modeled as attributes with customized attribute weights to represent their relevance to a known system fault and vice versa. Second, a tradeoff analysis is conducted by comparing the beliefs for different faults. The diagnosed faults with bigger beliefs within a predetermined valve would be considered as concurrent faults. To further improve the prediction accuracy, an optimization approach with the differential evolutionary algorithm as the optimization engine is designed with model constraints reflecting the complex correlations between the indicative factors and concurrent faults. A marine diesel engine fault diagnosis case is studied for validation. Case study results show that: first, both single and concurrent faults can be diagnosed with high accuracy by the proposed BRB prediction approach; second, the BRBs with optimized attributes weights has shown superior performance than with equal and fixed attribute weights; and third, the new BRB prediction approach outperforms the previous approach with multiple sub-BRBs as well as other approaches to achieve a higher modeling accuracy.
KW - Belief rule base (BRB) prediction
KW - concurrent fault diagnosis
KW - customized attributes weights
KW - tradeoff analysis
KW - uncertainty
UR - http://www.scopus.com/inward/record.url?scp=85102729409&partnerID=8YFLogxK
U2 - 10.1109/JSYST.2020.2991161
DO - 10.1109/JSYST.2020.2991161
M3 - 文章
AN - SCOPUS:85102729409
SN - 1932-8184
VL - 15
SP - 1179
EP - 1190
JO - IEEE Systems Journal
JF - IEEE Systems Journal
IS - 1
M1 - 9115865
ER -